AI · Communications
A WebRTC platform that listens, understands, and assists—using speech-to-text, sentiment analysis, purchase prediction, and live agent recommendations.
Role
Product owner, architecture, development
Context
Yabasi product
Period
Ongoing product
Platform
Web · WebRTC
The challenge
In a call centre, the decisive moments happen during the conversation: a customer hesitates, gets irritated, signals interest. Conventional systems record the call and analyse it afterwards, when it is too late to act. The platform had to understand a conversation while it is still running and help the agent in that moment.
The approach
Calls run over a WebRTC-based communication layer. The audio is transcribed continuously, and the transcript and the voice itself are analysed in parallel: sentiment, behaviour and the likelihood that the customer will buy. The results are turned into concrete suggestions and shown to the agent live. After the call, the system writes the summary and the report.
What was built
- 01
Real-time sentiment analysis
Tracks how the mood of the conversation develops, turn by turn.
- 02
Speech-to-text (ASR)
Continuous transcription as the basis for every further analysis.
- 03
Voice analytics & purchase prediction
Estimates purchase potential from what is said and how it is said.
- 04
Agent recommendation engine
Suggests next actions to the agent based on live behavioural data.
- 05
Automatic summaries & reports
Every call ends with a written summary and report, without manual work.
The outcome
The platform is in use at several call centres, where it supports agents during live conversations and takes over the documentation afterwards—making teams noticeably more efficient.
Next project